Update docs

Author: YanWenKun

README.adoc

@@ -117,19 +117,24 @@ Test results on Arc B580 (12G) are as follows:
 |Lumina 2
 |🆗 Good
 
+|FLUX.1 Krea
+|🆗 Good
+
 |FLUX.2 [Klein] 9B
 |🆗 Good
 
 |Z-Image Turbo
-|🆗 Good
+|🆗 Slow on first run. Image oversize may cause freeze.
 
 |Qwen Image 2512
-|🆗 Good
+|❌ Fail to load
 
 |===
 
 Current known issues:
 
+. Compatibility and performance regression. Using the current version of Intel GPU drivers (101.8626) with PyTorch (2.11.0 RC) results in worse text-to-image performance compared to the previous version. Consider using an older version or use the latest development version (PyTorch 2.12 nightly).
+
 . Once VRAM overflows, the program crashes or freezes and needs to be restarted. Adding the `--disable-smart-memory` parameter by default alleviates this issue.
 
 ** Using "Disable Smart Memory Management" will increase model loading time. If you only use a single model after launching, you can disable this option to save time.
@@ -138,8 +143,9 @@ Current known issues:
 
 ** Closing GPU-accelerated programs (e.g., browsers) can free up some VRAM. After closing the browser, the program continues to run, and you can check the generation progress in the log window.
 
-. In current version (PyTorch 2.10.0), the XPU performance on Windows is inferior than that on Linux.
-It's using more VRAM and is compatible with less models, although the inference speed is similar. WSL2 has not been tested.
+. For ComfyUI text-to-image generation specifically, XPU performance on Windows is comprehensively inferior to Linux, and this has not improved significantly over several versions.
+Conversely, on Linux, each version shows visible improvements, and the performance gap continues to widen.
+If you plan to use ComfyUI heavily on Intel GPUs, consider using a Linux system (Fedora is recommended first, followed by the latest Ubuntu).
 
 ** https://github.com/YanWenKun/ComfyUI-Docker/tree/main/xpu[Docker image for XPU]
 

README.zh.adoc

@@ -116,19 +116,24 @@ image::docs/screenshot-launcher.zh.webp["启动器截图"]
 |Lumina 2
 |🆗 正常
 
+|FLUX.1 Krea
+|🆗 正常
+
 |FLUX.2 [Klein] 9B
 |🆗 正常
 
 |Z-Image Turbo
-|🆗 正常
+|🆗 首次运行缓慢；图像尺寸过大易造成假死
 
 |Qwen Image 2512
-|🆗 正常
+|❌ 无法加载
 
 |===
 
 目前已知问题：
 
+. 兼容性与性能倒退。使用 PyTorch 2.11.0 RC 与英特尔显卡驱动 101.8626 ，文生图表现不如上一版。可以考虑使用旧版，或直接使用最新的开发版（PyTorch 2.12 nightly）。
+
 . 一旦显存溢出，程序即崩溃或假死，需要重新启动。默认添加参数 `--disable-smart-memory` 后有所缓解。
 
 ** 勾选“禁用智能内存管理”会使模型运行完即释放显存，不会一直缓存于显存中，但该设置会增加模型加载用时。如果本次程序启动后只使用单一模型，也可取消勾选。
@@ -137,8 +142,9 @@ image::docs/screenshot-launcher.zh.webp["启动器截图"]
 
 ** 关闭使用 GPU 加速的程序（比如浏览器）可释放一部分显存。关闭浏览器后程序仍然在运行，可在日志窗口查看生成进度。
 
-. 当前版本（PyTorch 2.10.0）下，XPU 在 Windows 的表现不如 Linux。
-在 Windows 上显存占用更高，兼容的模型更少，不过推理速度相近。WSL2 未测试。
+. 仅就 ComfyUI 文生图而言，XPU 在 Windows 下的性能表现全面不如 Linux，且历经数个版本仍未有明显改善。
+相反， 在 Linux 下每个版本都有可见提升，差距越拉越大。
+如打算在 Intel GPU 上重度使用 ComfyUI，建议考虑使用 Linux 系统（优先推荐 Fedora，其次推荐最新版 Ubuntu）。
 
 ** https://github.com/YanWenKun/ComfyUI-Docker/tree/main/xpu-cn[用于 XPU 的 ComfyUI Docker 镜像]